Forever Security traced an AI agent's attempted malware install to Downloads

Forever Security researchers Gal Weizman and Omer Samet reported on October 22 that they stopped an AI coding agent from installing a malicious npm package after tracing the instruction to an 'AGENTS.md' file in a customer's Downloads folder. The customer incident led the researchers to test attacks against OpenAI's Codex and Anthropic's Claude Code, which produced different results and limits, though Forever could not establish how the original file arrived on the machine.

Forever Security traced an AI agent's attempted malware install to Downloads Researchers Gal Weizman and Omer Samet say a customer incident led them to test attacks against Codex and Claude Code, with different results and limits. By Ryan Merket https://runtimewire.com/author/ryan-merket · Published Primary source: BleepingComputer https://www.bleepingcomputer.com/news/security/ai-coding-agents-hijacked-using-agentsmd-file-in-downloads-folder/ Why it matters The report connects a routine browser download to an agent's ability to change configuration and install software. It also shows why the attack claims need careful boundaries: Forever observed a customer incident and reproduced techniques in tests, but could not establish how the original file arrived. Forever Security https://forever.security/?ref=runtimewire says it stopped an AI coding agent from installing a malicious npm package after tracing the instruction to an 'AGENTS.md' file in a customer's Downloads folder. The October 22nd report by researchers Gal Weizman and Omer Samet https://forever.security/blog/hijacking-ai-agents-via-downloads-folder?ref=runtimewire documents a path from an ordinary downloaded text file to commands run by an agent with access to the customer's machine.… When you purchase through links in our articles, we may earn a small commission https://runtimewire.com/affiliate-monetization-standards . This doesn’t affect our editorial independence.